Breast Region Segmentation being Convolutional Neural Network in Dynamic Contrast Enhanced MRI

Area of Science:

• Medical Imaging
• Artificial Intelligence in Radiology

Background:

• Breast density and background parenchymal enhancement (BPE) are linked to breast cancer risk.
• Accurate breast segmentation is essential for quantitative analysis of these factors.
• Convolutional neural networks (CNNs) show promise for medical image segmentation.

Purpose of the Study:

• To employ a CNN for segmenting breast regions in transverse fat-suppressed breast dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI).
• To establish a reliable method for quantitative analysis of breast density and BPE.

Main Methods:

• Image normalization and random dataset splitting into training, validation, and test sets.
• Training a 2-D U-Net model on the training set and optimizing with the validation set.
• Postprocessing segmentation results, including scar identification, and employing 5-fold cross-validation and data augmentation due to dataset limitations.

Main Results:

• The 2-D U-Net model achieved high accuracy in breast region segmentation.
• Mean Dice Similarity Coefficient (DSC) of 97.44%, Dice Difference Coefficient (DDC) of 5.11%, and root-mean-square distance of 1.25 pixels were obtained.
• The method demonstrated effectiveness in segmenting breast tissue for further analysis.

Conclusions:

• CNN-based segmentation, specifically using 2-D U-Net, is a viable and accurate method for breast region segmentation in DCE-MRI.
• This approach facilitates quantitative analysis of breast density and BPE, contributing to breast cancer risk assessment.
• The study highlights the potential of AI in improving radiological assessments for cancer risk.